November 14, 2025
Jensen wants the whole kitchen
Nvidia is gearing up to sell servers instead of just GPUs and components
Nvidia wants to sell the whole AI box — hype, Microsoft playbook, or ‘AI factories’
TLDR: JP Morgan says Nvidia may ship fully built AI server guts with the Vera Rubin platform, grabbing more control and profit. Commenters split between “smart move” and “monopoly/PR stunt,” with jokes about selling AI itself and bundling ChatGPT, highlighting fears of lock-in and fewer choices in a booming AI hardware market.
Nvidia is reportedly planning to take over the whole server build for its next‑gen Vera Rubin platform, shipping pre‑assembled “L10” trays — basically the guts of an AI server: chips, memory, networking, and liquid cooling — straight to partners. Cue comment chaos. One jokester quipped “Soon Nvidia will sell AI itself,” while another asked if this is even new because Nvidia already sells DGX servers. The mood? Split between “smart, faster rollouts” and “monopoly vibes.”
Skeptics say Nvidia is already supply constrained, so taking on more of the build could slow things and looks like stock‑boosting PR. Others see the classic Microsoft playbook: lock in the stack, maybe even bundle ChatGPT. One snark: forget servers, it’s “AI factories” now. Stakes: if Nvidia ships ~90% of the server as a tested module, partners are reduced to rack assembly while Nvidia grabs more profit and power. Fans cheer simplicity and faster rollout; critics worry about fewer choices, tighter lock‑in, and Jensen owning the whole kitchen. Drama: loud.
Key Points
- •J.P. Morgan reports Nvidia may sell fully assembled L10 compute trays with its VR200 (Vera Rubin) platform.
- •L10 trays would include CPU, GPUs, memory, NICs, power delivery, midplane interfaces, and liquid-cooling cold plates.
- •This would reduce ODM server design work to rack-level integration and operational tasks.
- •Nvidia previously offered partial integration with the GB200’s Bianca board (L7–L8).
- •The shift could accelerate VR200 ramp and lower costs via EMS scale, with Foxconn, Quanta, and Wistron cited speculatively.